Liver Segmentation in Computed Tomography Images Using Transformers, Inception Module and U-Net

Abstract

Hepatic surgery requires the segmentation of the liver from computed tomography (CT) images. Fully automated approaches are required since manual segmentation requires much time and labor. This paper proposes a deep learning model with the basic architecture U-Net to segment the liver, including a transformer and an inception modules. The Dice coefficient on the test phase was 84.8% and the Jaccard coefficient was 63.6% on 332 CT images. This work shows the usefulness of the transformers in the liver segmentation.